JP6908134B2 - Hypothesis generator, hypothesis generation method, and program - Google Patents

Description

The present invention, for generating a hypothesis by hypothetical reasoning technology relates hypothesis generator and hypothesis generation method further relates to a program for realizing these.

Hypothetical reasoning is to make a hypothetical reasoning that does not contradict the events observed when a certain thing occurs and the reasoning knowledge (for example, common sense). In general hypothesis reasoning, observations are considered to be given, but in order to improve and verify the hypothesis, a device that executes hypothesis reasoning may actively acquire observations from the outside (for example,). See Patent Documents 1 and 2).

FIG. 10 is a block diagram showing a configuration of a device that executes conventional hypothesis inference. In the example of FIG. 10, the hypothesis generator has a function of actively acquiring observations. Further, as shown in FIG. 10, the hypothesis generation device includes a hypothesis generation unit, a hypothesis verification unit, and an inference knowledge storage unit. Further, the hypothesis generation device is connected to an observation acquisition destination (observation acquisition source) outside the device.

Further, in the hypothesis generation device, the hypothesis generation unit generates a hypothesis. Specifically, the hypothesis generation unit first accepts observations that are first-order predicate logic expressions in which all terms have values. Next, the inference generator uses the knowledge data represented by the first-order predicate logical expression including the logical inclusion among the first-order predicate logical expressions, and from the received observation to the possible consequences, the first-order predicate logical expression , Outputs the hypothetical inference result composed of the cost of evaluating the badness of those first-order predicate logic expressions.

The hypothesis verification unit identifies and identifies first-order predicate logic expressions that include terms (variables) that have values obtained from observations and terms that do not have values, among the first-order predicate logic expressions that make up the hypothesis inference results. The observation in which the value is included in the term that does not have the value in the first-order predicate logic expression is acquired from the observation acquisition source. After that, the hypothesis generation unit verifies the hypothesis inference result output earlier by using the observation acquired by the hypothesis verification unit, and corrects it if necessary.

In addition, Non-Patent Documents 1 to 3 specifically disclose research on hypothesis generation processing performed by the above-mentioned hypothesis generation unit.

Japanese Unexamined Patent Publication No. 8-122398 Japanese Unexamined Patent Publication No. 2016-91039

Inoue, Naoya, and Kentaro Inui. ILP-Based Reasoning for WeightedAbduction. Plan, Activity, and Intent Recognition. 2011. Inoue, Naoya, et al. Coreference Resolution with ILP-based WeightedAbduction. COLING. 2012. Kazeto Yamamoto, Naoya Inoue, Kentaro Inui, Yuki Arase and Junichi Tsujii. Boosting the Efficiency of First-order Abductive Reasoning Using Pre-estimated Relatedness between Predicates. International Journal of MachineLearning and Computing, Vol. 5, No. 2, pp. 114-120 . April 2015.

By the way, in the conventional hypothesis generation device shown in FIG. 10, the hypothesis verification unit needs to acquire observations from an external observation acquisition source in order to verify the hypothesis inference result derived by the hypothesis generation unit. On the other hand, it is necessary to minimize the number of observations acquired from the observation acquisition source, which is required for the hypothesis generation unit to output the hypothesis inference result having a predetermined cost.

However, in the device shown in FIG. 10, since the acquisition of observations from an external observation acquisition source at a certain point in time is performed only based on the hypothesis inference results derived at the same point, it tends to be redundant and the observations are observed. It is difficult to reduce the number of acquisitions. This point will be specifically described below.

First, the hypothesis inference result output from the hypothesis generation unit changes non-monotonically when the observation output from the hypothesis verification unit is added. Therefore, the observations required for hypothetical inference results will differ from time to time. Therefore, in order to obtain a specific hypothesis inference result, that is, a hypothesis inference result having a cost below a specified threshold through a small number of observation acquisitions, the hypothesis inference result (particularly cost) output from the hypothesis generation unit is required. It is necessary to predict how it will change with the addition of observations and acquire observations from the observation acquisition source. However, such a prediction is extremely difficult, and as a result, as described above, the acquisition of observations from the observation acquisition source becomes redundant.

An example of an object of the present invention is to provide a hypothesis generator, a hypothesis generation method, and a program capable of solving the above-mentioned problems and suppressing redundancy in the acquisition of observations necessary for verification of hypothesis inference results. be.

In order to achieve the above object, the hypothesis generator in one aspect of the present invention is
When a query by a first-order predicate logical expression including a term without a value is accepted, the observation model that outputs the observation corresponding to the query according to the set rule,
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
An observation acquisition unit that acquires the observation to which the cost is given, and
Knowledge of accepting and accepting observations that are first-order predicate logical expressions in which all terms have values, and the observations acquired by the observation acquisition unit, expressed by first-order predicate logical expressions including logical inclusion. A hypothesis generation unit that generates hypothetical inference results, which is composed of a first-order predicate logical expression that is applied to data and leads to a possible outcome, and a cost that indicates the evaluation of the hypothesis by the first-order predicate logical expression.
Is equipped with
It is characterized by that.

Further, in order to achieve the above object, the hypothesis generation method in one aspect of the present invention is:
(A) When an inquiry by a first-order predicate logical expression including a term having no value is accepted, an observation model corresponding to the inquiry is output according to the set rule.
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
To obtain the observation with the cost, and
(B) Accepting an observation that is a first-order predicate logical expression in which all terms have values, and the observation obtained in the step (a) above, the first-order predicate logical expression including logical inclusion. A step of generating a hypothetical inference result composed of a first-order predicate logical expression leading to a possible consequence and a cost indicating evaluation of a hypothesis by the first-order predicate logical expression by applying it to the knowledge data represented by. When,
Have,
It is characterized by that.

Further, in order to achieve the above object, the program in one aspect of the present invention is:
On the computer
(A) When an inquiry by a first-order predicate logical expression including a term having no value is accepted, an observation model corresponding to the inquiry is output according to the set rule.
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
To obtain the observation with the cost, and
(B) Accepting an observation that is a first-order predicate logical expression in which all terms have values, and the observation obtained in the step (a) above, the first-order predicate logical expression including logical inclusion. A step of generating a hypothetical inference result composed of a first-order predicate logical expression leading to a possible consequence and a cost indicating evaluation of a hypothesis by the first-order predicate logical expression by applying it to the knowledge data represented by. When,
Ru is running, characterized in that.

As described above, according to the present invention, it is possible to suppress the redundancy of the acquisition of observations necessary for the verification of the hypothesis inference result.

FIG. 1 is a block diagram showing a schematic configuration of a hypothesis generator according to an embodiment of the present invention. FIG. 2 is a block diagram showing a specific configuration of the hypothesis generator according to the embodiment of the present invention. FIG. 3 is a flow chart showing an operation at the time of updating the observation acquisition plan of the hypothesis generation device 100 according to the embodiment of the present invention. FIG. 4 is a flow chart showing the operation of the hypothesis generation device 100 during the hypothesis generation processing according to the embodiment of the present invention. FIG. 5 is a block diagram showing an example of a computer that realizes the hypothesis generator according to the embodiment of the present invention. FIG. 6 is a diagram showing an embodiment of steps A1 to A4 shown in FIG. FIG. 7 is a diagram showing an embodiment of step A5 shown in FIG. FIG. 8 is a diagram showing an embodiment of steps B1 to B3 shown in FIG. FIG. 9 is a diagram showing an embodiment of step B4 shown in FIG. FIG. 10 is a block diagram showing a configuration of a device that executes conventional hypothesis inference.

(Embodiment)
Hereinafter, the hypothesis generation device, the hypothesis generation method, and the program according to the embodiment of the present invention will be described with reference to FIGS. 1 to 5.

[Device configuration]
First, the schematic configuration of the hypothesis generation device according to the present embodiment will be described with reference to FIG. FIG. 1 is a block diagram showing a schematic configuration of a hypothesis generator according to an embodiment of the present invention.

The hypothesis generation device 100 in the present embodiment shown in FIG. 1 is a device that generates a hypothesis by using a hypothesis inference technique. As shown in FIG. 1, the hypothesis generation device 100 includes an observation acquisition unit 30 and a hypothesis generation unit 40.

Further, the hypothesis generation device 100 uses the knowledge data represented by the first-order predicate logic expression including the logical inclusion and the observation acquisition plan. The observation acquisition plan is expressed by a first-order predicate logic formula, and registers the observation to be inquired and the cost to be obtained in the future.

The observation acquisition unit 30 inputs an observation acquisition plan into the observation model 50 and acquires observations to which a cost has been added. The observation model 50 is a model that receives an inquiry by a first-order predicate logic expression including a term having no value and outputs an observation corresponding to the inquiry according to a set rule. The observation model 50 is actually held in an external device of the hypothesis generation device 100.

The hypothesis generation unit 40 accepts observations that are first-order predicate logical expressions in which all terms have values, and applies the received observations and the observations acquired by the observation acquisition unit 30 to the knowledge data. The hypothesis generation unit 40 is composed of a first-order predicate logic formula that leads to a possible consequence by applying these two observations to the knowledge data, and a cost that indicates the evaluation of the hypothesis by this first-order predicate logic formula. , Generate hypothesis inference results.

Further, in the present embodiment, the "cost indicating the evaluation of the hypothesis by the first-order predicate logic formula" is, in other words, the cost of evaluating the badness in the first-order predicate logic formula constituting the hypothesis inference result (the following). See references). "Badness" means low probability and explanatoryity of hypothetical inference results composed of first-order predicate logical expressions, and probability and explanatoryness use weights embedded in knowledge data. Is calculated.
[References]
Inoue, Naoya, and Kentaro Inui. "Large-scale cost-basedabduction in full-fledged first-order predicate logic with cutting plane inference." Logics in Artificial Intelligence. Springer, Berlin, Heidelberg, 2012. 281-293.

As described above, in the present embodiment, when the hypothesis inference result is generated from the received observations and knowledge data, the observations acquired according to the observation acquisition plan are added. Therefore, according to the present embodiment, redundancy is suppressed in the acquisition of observations necessary for verifying the hypothesis inference result.

Subsequently, the configuration of the hypothesis generation device 100 in the present embodiment will be described more specifically with reference to FIG. FIG. 2 is a block diagram showing a specific configuration of the hypothesis generator according to the embodiment of the present invention.

As shown in FIG. 2, the hypothesis generation device 100 in the present embodiment includes the inference knowledge storage unit 10, the observation acquisition plan storage unit 20, and the observation output in addition to the observation acquisition unit 30 and the hypothesis generation unit 40 described above. A unit 60, a simulation execution unit 70, and an observation acquisition plan update unit 80 are provided.

Of these, the inference knowledge storage unit 10 stores the above-mentioned knowledge data. The observation acquisition plan storage unit 20 stores the above-mentioned observation acquisition plan.

When the observation output unit 60 receives an inquiry by a first-order predicate logic expression including a term having no value, the observation output unit 60 outputs an observation corresponding to the inquiry according to a rule used by the observation model 50 or a rule different from this rule. As a rule different from the rule used by the observation model 50, there is a rule similar to the rule used by the observation model 50.

The similar rule is a rule obtained by approximating the input / output correspondence of the observation model 50. When an example of the observation sequence output from the observation model 50 is obtained for a certain query sequence, for example, based on supervised learning, it is possible to obtain a rule similar to the rule used by the observation model 50. You can.

The simulation execution unit 70 causes the observation acquisition unit 30 to input the observation acquisition plan to the observation output unit 60 and acquire the observation output by the input. Further, the simulation execution unit 70 gives the hypothesis generation unit 40 the observations acquired by the observation acquisition unit 30 and the observations as samples, and causes the hypothesis inference result to be generated. Further, the simulation execution unit 70 causes the observation acquisition unit 30 to acquire observations and causes the hypothesis generation unit 40 to generate the hypothesis inference result until the cost for constructing the generated hypothesis inference result becomes equal to or less than a preset threshold value. ..

The observation acquisition plan update unit 80 updates the observation acquisition plan using the cost included in the hypothesis inference result generated by the simulation execution unit 70.

Further, in the present embodiment, the observation acquisition unit 30 acquires the observation to which the cost is given in accordance with the rule stipulating that the observation with the lowest cost in the observation acquisition plan is applied to the observation model 50. ..

[Device operation]
Next, the operation of the hypothesis generator 100 according to the embodiment of the present invention will be described with reference to FIGS. 3 and 4. In the following description, FIGS. 1 and 2 will be referred to as appropriate. Further, in the present embodiment, the hypothesis generation method is implemented by operating the hypothesis generation device 100. Therefore, the description of the hypothesis generation method in the present embodiment is replaced with the following description of the operation of the hypothesis generation device 100.

First, the update process of the observation acquisition plan will be described with reference to FIG. FIG. 3 is a flow chart showing an operation at the time of updating the observation acquisition plan of the hypothesis generation device 100 according to the embodiment of the present invention.

First, as shown in FIG. 3, the simulation execution unit 70 receives input from the outside of the threshold of the cost constituting the hypothesis inference result generated by the simulation and the observation as a sample (step A1).

Next, the simulation execution unit 70 causes the observation acquisition unit 30 to input the observation acquisition plan to the observation output unit 60 and acquire the observation output by the input (step A2). As a result, the observation acquisition unit 30 outputs the acquired observation to the simulation execution unit 70.

Next, the simulation execution unit 70 gives the hypothesis generation unit 40 the observations acquired by the observation acquisition unit 30 in step A2 and the observations as samples, and causes the hypothesis inference result to be generated (step A3).

Next, the simulation execution unit 70 determines whether or not the cost constituting the hypothesis inference result generated in step A3 is equal to or less than a preset threshold value (step A4). Then, if the cost is not equal to or less than the threshold value as a result of the determination in step A4, the simulation execution unit 70 executes steps A2 and A3 again.

On the other hand, if the cost is equal to or less than the threshold value as a result of the determination in step A4, the simulation execution unit 70 instructs the observation acquisition plan update unit 80 to update the observation acquisition plan. As a result, the observation acquisition plan update unit 80 updates the observation acquisition plan in the observation acquisition plan storage unit 20 using the cost included in the hypothesis inference result generated in step A3 (step A5).

Subsequently, the hypothesis generation process will be described with reference to FIG. FIG. 4 is a flow chart showing the operation of the hypothesis generation device 100 during the hypothesis generation processing according to the embodiment of the present invention.

As shown in FIG. 4, first, the hypothesis generation unit 40 accepts the input of the observation which is the first-order predicate logic expression in which all the terms have values (step B1).

Next, the observation acquisition unit 30 acquires the observation acquisition plan from the observation acquisition plan storage unit 20 (step B2). Next, the observation acquisition unit 30 inputs the observation acquisition plan received in step B2 into the observation model 50, and acquires the observation to which the cost is added (step B3). Further, the observation acquisition unit 30 outputs the acquired observation to the hypothesis generation unit 40.

Next, the hypothesis generation unit 40 applies the observation received in step B1 and the observation acquired in step B3 to the knowledge data stored in the inference knowledge storage unit 10 to execute hypothesis inference. The hypothesis inference result obtained as a result of the execution is output (step B4).

[Effect in the embodiment]
As described above, in the present embodiment, when the hypothesis inference result is generated, the observations acquired according to the observation acquisition plan are added. In addition, the observation acquisition plan has been updated in advance by simulation so that the cost is below the threshold value. Therefore, even if the hypothesis inference result output from the hypothesis generation unit 40 changes non-monotonically, according to the present embodiment, it is redundant until the hypothesis inference result having a cost equal to or less than the specified threshold value is derived. Observation acquisition is reduced.

[program]
The program according to the present embodiment may be any program that causes a computer to execute steps A1 to A5 shown in FIG. 3 and steps B1 to B4 shown in FIG. By installing this program on a computer and executing it, the hypothesis generation device 100 and the hypothesis generation method according to the present embodiment can be realized. In this case, the computer processor functions as an observation acquisition unit 30, a hypothesis generation unit 40, an observation output unit 60, a simulation execution unit 70, and an observation acquisition plan update unit 80, and performs processing.

Further, in the present embodiment, the inference knowledge storage unit 10 and the observation acquisition plan storage unit 20 are realized by storing the data files constituting them in a storage device such as a hard disk provided in the computer.

Further, the program in the present embodiment may be executed by a computer system constructed by a plurality of computers. In this case, for example, each computer may function as one of the observation acquisition unit 30, the hypothesis generation unit 40, the observation output unit 60, the simulation execution unit 70, and the observation acquisition plan update unit 80, respectively. Further, the inference knowledge storage unit 10 and the observation acquisition plan storage unit 20 may be constructed on a computer different from the computer that executes the program in the present embodiment.

[Physical configuration]
Here, a computer that realizes the hypothesis generator 100 by executing the program according to the present embodiment will be described with reference to FIG. FIG. 5 is a block diagram showing an example of a computer that realizes the hypothesis generator according to the embodiment of the present invention.

As shown in FIG. 5, the computer 110 includes a CPU 111, a main memory 112, a storage device 113, an input interface 114, a display controller 115, a data reader / writer 116, and a communication interface 117. Each of these parts is connected to each other via a bus 121 so as to be capable of data communication. The computer 110 may include a GPU (Graphics Processing Unit) or an FPGA (Field-ProgrammableGate Array) in addition to the CPU 111 or in place of the CPU 111.

The CPU 111 expands the programs (codes) of the present embodiment stored in the storage device 113 into the main memory 112 and executes them in a predetermined order to perform various operations. The main memory 112 is typically a volatile storage device such as a DRAM (Dynamic Random Access Memory). Further, the program according to the present embodiment is provided in a state of being stored in a computer-readable recording medium 120. The program in the present embodiment may be distributed on the Internet connected via the communication interface 117.

Further, specific examples of the storage device 113 include a semiconductor storage device such as a flash memory in addition to a hard disk drive. The input interface 114 mediates data transmission between the CPU 111 and an input device 118 such as a keyboard and mouse. The display controller 115 is connected to the display device 119 and controls the display on the display device 119.

The data reader / writer 116 mediates the data transmission between the CPU 111 and the recording medium 120, reads the program from the recording medium 120, and writes the processing result in the computer 110 to the recording medium 120. The communication interface 117 mediates data transmission between the CPU 111 and another computer.

Specific examples of the recording medium 120 include a general-purpose semiconductor storage device such as CF (Compact Flash (registered trademark)) and SD (Secure Digital), a magnetic recording medium such as a flexible disk, or a CD-. Examples include optical recording media such as ROM (Compact DiskRead Only Memory).

The hypothesis generation device 100 in the present embodiment can also be realized by using hardware corresponding to each part instead of the computer in which the program is installed. Further, the hypothesis generator 100 may be partially realized by a program and the rest may be realized by hardware.

Next, an example of the hypothesis generation device 100 according to the present embodiment will be described with reference to FIGS. 6 to 9. In addition, in the following, each step shown in FIGS. 3 and 4 described above will be described.

[Steps A1 to A4]
FIG. 6 is a diagram showing an embodiment of steps A1 to A4 shown in FIG. As shown in FIG. 6, the simulation execution unit 70 uses the inference knowledge storage unit 10, the observation acquisition plan storage unit 20, the observation acquisition unit 30, the hypothesis generation unit 40, and the observation output unit 60 to make a hypothesis. Output the inference result (observation acquisition plan execution result).

In this embodiment, as shown in FIG. 6, it is assumed that the observation input to the simulation execution unit 70 is "C (Taro, y) ^{$ 10} ". Also, in this observation, arguments that start with a lowercase letter are variables, and arguments that start with an uppercase letter are constants. In addition, the "$ 10" on the right shoulder of the literal that indicates the observation is the cost of that literal.

Further, as shown in FIG. 6, the inference knowledge storage unit 10 contains the knowledge data “A (x, y) ^0.8 → C (x, y)” and “B (x, y) ² → C (x, y)”. y) ”is stored. These knowledge data represent weighted inference rules in weighted hypothesis inference (see Non-Patent Documents 1 to 3). The number on the right shoulder of the literal showing the knowledge data represents the weight.

As shown in FIG. 6, the observation output unit 60 has two types of rules, “If A (Taro, y) is queried, return null” and “If B (Taro, y) is queried, returnB (Taro, Jiro) ^{$ 0.} The observation is output according to. These rules state that "when a literal (eg, A (Taro, y)) is entered into the observation model as a query, it returns a specific value (eg, null)".

As shown in FIG. 6, the observation acquisition plan storage unit 20 has "[Next: A (Taro, y), CCost: 0]" and "[Next: B (Taro, y), CCost: 0]" as the observation acquisition plan. : 0] ”is stored. Further, in the example of FIG. 6, the observation acquisition plan is represented as a dictionary, "Next" includes a first-order predicate logic expression to be queried to the observation model 50, and "C Cost" is a first-order predicate logic to be queried. The future cost of the formula is included as an assessment.

Under the above conditions, the simulation execution unit 70 performs steps A2 (observation acquisition simulation 1) and A3 (hypothesis generation simulation 1) according to the observation acquisition plan “[Next: A (Taro, y), CCost: 0]”. Execute. Further, the simulation execution unit 70 also executes steps A2 (observation acquisition simulation 2) and A3 (hypothesis generation simulation 2) according to “[Next: B (Taro, y), CCost: 0]”.

Specifically, the observation acquisition unit 30 sends Query: A (Taro, y) to the observation output unit 60 according to the observation acquisition plan [Next: A (Taro, y), CCost: 0] according to the instruction of the simulation execution unit 70. Input y) (observation acquisition simulation 1). As a result, the observation output unit 60 accepts this observation acquisition plan as an inquiry and returns "null" based on the included rules (upper block of FIG. 7). Then, the observation acquisition unit 30 acquires "null" as an observation and outputs this to the simulation execution unit 70.

Next, the hypothesis generation unit 40 uses the observation "null", the knowledge data included in the inference knowledge storage unit 10, and the input observation "C (Taro, y) ^{$ 10} " according to the instruction of the simulation execution unit 70. Hypothesis inference is performed and the hypothesis inference result is output (hypothesis generation simulation 1). This hypothesis inference result includes the cost of performing weighted hypothesis inference (“$ 8” in the example of FIG. 6).

Further, in this embodiment, the cost is calculated according to the calculation method described in the above-mentioned reference. Specifically, the hypothesis generation unit 40 applies ^{"C (Taro, y) $ 10} " to the knowledge data "A (x, y) ^0.8 → C (x, y)" included in the inference knowledge storage unit 10. , Perform a backward chain and hypothesize "C (Taro, y) ^{$ 08".} The result is a cost of $ 8.

Subsequently, the observation acquisition unit 30 inputs Query: B (Taro, y) to the observation output unit 60 according to the remaining observation acquisition plan [Next: B (Taro, y), CCost: 0] (observation acquisition). Simulation 2). As a result, the observation output unit 60 accepts this observation acquisition plan as an inquiry and returns "B (Taro, Jiro)" based on the included rules. Then, the observation acquisition unit 30 acquires "B (Taro, Jiro)" as an observation, and outputs this to the simulation execution unit 70.

Next, the hypothesis generation unit 40 receives the observation "B (Taro, Jiro)" and the knowledge data "B (x, y) ² → C (x, x,) included in the reasoning knowledge storage unit 10 according to the instruction of the simulation execution unit 70. Hypothesis inference is performed using "y)" and the input observation "C (Taro, y) ^{$ 10} ", and the hypothesis inference result is output (hypothesis generation simulation 2). The hypothesis inference result in this case also includes the cost of performing the weighted hypothesis inference (“$ 0” in the example of FIG. 6).

Specifically, the hypothesis generation unit 40 applies ^{"C (Taro, y) $ 10} " to the knowledge data "B (x, y) ² → C (x, y)" included in the inference knowledge storage unit 10. , Perform a backward chain and hypothesize "C (Taro, y) ^{$ 20".} As a result, the cost will be $ 20, but in hypothesis generation simulation 2, the cost will be unified with the observation acquisition plan [Next: B (Taro, y), CCost: 0] ", so the final cost will be. It will be $ 0.

In addition, in the observation acquisition simulation 1 and the hypothesis generation simulation 1, the observation acquisition unit 30 also inputs Query: B (Taro, y) to the observation output unit 60. In this case, the hypothesis generator calculates the cost as "$ 0". Further, in the observation acquisition simulation 2 and the hypothesis generation simulation 2, the observation acquisition unit 30 also inputs Query: A (Taro, y) to the observation output unit 60. In this case as well, the hypothesis generator calculates the cost as "$ 0".

As a result of the above, as shown in FIG. 6, as a result of steps A1 to A4, the cost (observation acquisition plan execution result) included in the hypothesis inference result can be obtained. The simulation execution unit 70 outputs the obtained observation acquisition plan execution result to the observation acquisition plan update unit 80.

[Step A5]
FIG. 7 is a diagram showing an embodiment of step A5 shown in FIG. As shown in FIG. 7, the observation acquisition plan update unit 80 receives the observation acquisition plan execution result as an input, and updates the observation acquisition plan based on the received observation acquisition plan execution result.

As shown in FIG. 7, the observation acquisition plan update unit 80 updates the CCost included in the observation acquisition plan based on the observation acquisition plan execution result obtained in the example of FIG. In this embodiment, the cumulative value (total) of the costs (total cost) included in the observation acquisition plan execution result is treated as CCost, and the observation acquisition plan is updated.

Specifically, CCost corresponding to "Next: A (Taro, y)" is the sum of "$ 8" calculated by hypothesis generation simulation 1 and "$ 0" calculated by hypothesis generation simulation 2 and $ 8. Become. The CCost corresponding to "Next: B (Taro, y)" is the sum of "$ 0" calculated by the hypothesis generation simulation 1 and "$ 0" calculated by the hypothesis generation simulation 2, which is $ 0.

[Steps B1 to B3]
FIG. 8 is a diagram showing an embodiment of steps B1 to B3 shown in FIG. As shown in FIG. 8, the hypothesis generation unit 40 accepts an input of observation, which is a first-order predicate logic expression in which all terms have values. The observation acquisition unit 30 receives the observation acquisition plan as an input, acquires observations from the observation model 50 based on the observation acquisition plan, and outputs the acquired observations.

Specifically, the observation acquisition unit 30 selects the observation acquisition plan having the lowest CCost (in this example, [Next: B (Taro, y), CCost: 0]), and based on the selected observation acquisition plan. , Make an inquiry to the observation model 50 (in this embodiment, B (Taro, y) is inquired).

Further, in this embodiment, as shown in FIG. 8, B (Taro, y) ^{$ 0} is acquired from the observation model 50 as a result of the inquiry. The observation acquisition unit 30 outputs the acquired observation and the observation C (Taro, y) ^{$ 10} previously input and held internally. In this embodiment, it is assumed that the observation model 50 outputs observations in accordance with the same rules as those included in the observation output unit 60.

[Step B4]
FIG. 9 is a diagram showing an embodiment of step B4 shown in FIG. As shown in FIG. 9, the hypothesis generation unit 40 receives the observation received in step B1 and the observation output in step B3 as inputs, and performs hypothesis inference based on the received observation and knowledge data. Output the hypothesis reasoning result. As shown in FIG. 9, the cost of the obtained hypothesis inference result is derived from the hypothesis inference result having a cost equal to or less than a preset threshold value (“Total cost $ 0” in this embodiment).

[Comparison with conventional products]
By the way, when observations are acquired using the conventional hypothesis generator shown in FIG. 10, in order to derive a hypothesis inference result having a cost below a preset threshold (for example, total cost $ 0), It is necessary to acquire observations from the observation acquisition source (observation model) at least twice. This point will be described below.

First, in the conventional hypothesis generator shown in FIG. 10, the following two methods have been proposed. The first method is a method in which the hypothesis generation unit generates a hypothesis inference result with the minimum cost based on the observation and knowledge data, and queries the observation acquisition source for the first-order predicate logical expression included in the hypothesis inference result. be. The second method is a method of obtaining observations by an analysis applicable to all hypothesis inference results derived by the hypothesis generation unit based on observations and knowledge data. The case of acquiring observations based on the observations and knowledge data shown in FIG. 6 will be examined for the above two methods.

First, considering the case of the first method, the hypothesis that minimizes the cost in this embodiment is A (Taro, y) ^{$ 8} , and the hypothesis verification unit ^{obtains A (Taro, y) $ 8} as the observation acquisition source. Contact. In this case, null is returned from the observation source. Assuming that this "null" and C (Taro, y) ^{$ 10} are observed, the cost included in the hypothesis inference result when input to the hypothesis generation unit is $ 8, which has not reached $ 0.

Next, considering the case of the second method, the hypothesis generation unit outputs A (Taro, y) ^{$ 0.8} and B (Taro, y) ^{$ 20} , respectively, as hypothesis inference results. Based on this result, the hypothesis verification unit treats A (Taro, y) and B (Taro, y) as candidates for inquiry to the observation acquisition source. Then, in the second method, since it is not possible to determine which of the plurality of inquiry candidates is to be inquired, the hypothesis verification unit needs to make an inquiry to the observation acquisition source twice.

Based on the above, according to the present embodiment, it is possible to reduce redundant observation acquisition until a hypothesis inference result having a cost equal to or less than a preset threshold value is derived, as compared with the case of using a conventional hypothesis generator. Understand.

A part or all of the above-described embodiments can be expressed by the following descriptions (Appendix 1) to (Appendix 12), but the present invention is not limited to the following description.

(Appendix 1)
When a query by a first-order predicate logical expression including a term without a value is accepted, the observation model that outputs the observation corresponding to the query according to the set rule,
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
An observation acquisition unit that acquires the observation to which the cost is given, and
Knowledge of accepting and accepting observations that are first-order predicate logical expressions in which all terms have values, and the observations acquired by the observation acquisition unit, expressed by first-order predicate logical expressions including logical inclusion. A hypothesis generation unit that generates hypothetical inference results, which is composed of a first-order predicate logical expression that is applied to data and leads to a possible outcome, and a cost that indicates the evaluation of the hypothesis by the first-order predicate logical expression.
Is equipped with
A hypothesis generator characterized by this.

(Appendix 2)
The hypothesis generator according to Appendix 1,
When an inquiry by a first-order predicate logical expression including a term having no value is received, an observation output unit that outputs an observation corresponding to the inquiry according to the rule used by the observation model or a rule different from the rule. ,
The observation acquisition unit is made to input the observation acquisition plan to the observation output unit and the observations output by the input are acquired, and the hypothesis generation unit is made to acquire the observations and samples. A simulation execution unit that gives observations and generates hypothesis inference results,
An observation acquisition plan update unit that updates the observation acquisition plan using the cost included in the hypothesis inference result generated by the simulation execution unit.
A hypothesis generator characterized by further comprising.

(Appendix 3)
The hypothesis generator described in Appendix 2,
The simulation execution unit causes the observation acquisition unit to acquire observations and causes the hypothesis generation unit to generate hypothesis inference results until the cost of constructing the generated hypothesis inference result becomes equal to or less than a preset threshold value.
A hypothesis generator characterized by this.

(Appendix 4)
The hypothesis generator according to any one of Supplementary notes 1 to 3.
The observation acquisition unit acquires the observation with the cost in accordance with the rule stipulating that the observation with the lowest cost in the observation acquisition plan is applied to the observation model.
A hypothesis generator characterized by this.

(Appendix 5)
(A) When an inquiry by a first-order predicate logical expression including a term having no value is accepted, an observation model corresponding to the inquiry is output according to the set rule.
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
To obtain the observation with the cost, and
(B) Accepting an observation that is a first-order predicate logical expression in which all terms have values, and the observation obtained in the step (a) above, the first-order predicate logical expression including logical inclusion. A step of generating a hypothetical inference result composed of a first-order predicate logical expression leading to a possible consequence and a cost indicating evaluation of a hypothesis by the first-order predicate logical expression by applying it to the knowledge data represented by. When,
Have,
A hypothesis generation method characterized by this.

(Appendix 6)
The hypothesis generation method described in Appendix 5,
(C) When an inquiry by a first-order predicate logic expression including a term having no value is received, an observation corresponding to the inquiry is output according to the rule used by the observation model or a rule different from the rule. When,
(D) When the observation acquisition plan is input in the step (c), the observations output by the observation acquisition plan are acquired, and the acquired observations and the sample observations are used for hypothesis inference. Steps and steps to generate results,
(E) A step of updating the observation acquisition plan using the cost included in the hypothesis inference result generated in the step (d).
A hypothesis generation method characterized by further having.

(Appendix 7)
The hypothesis generator according to Appendix 6,
Step (d) is executed until the cost of constructing the generated hypothetical inference result is less than or equal to a preset threshold.
A hypothesis generation method characterized by this.

(Appendix 8)
The hypothesis generation method according to any one of Appendix 5 to 7.
In the step (a), the observation with the cost is acquired in accordance with the rule stipulating that the observation with the lowest cost in the observation acquisition plan is applied to the observation model.
A hypothesis generation method characterized by this.

(Appendix 9)
On the computer
(A) When an inquiry by a first-order predicate logical expression including a term having no value is accepted, an observation model corresponding to the inquiry is output according to the set rule.
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
To obtain the observation with the cost, and
(B) Accepting an observation that is a first-order predicate logical expression in which all terms have values, and the observation obtained in the step (a) above, the first-order predicate logical expression including logical inclusion. A step of generating a hypothetical inference result composed of a first-order predicate logical expression leading to a possible consequence and a cost indicating evaluation of a hypothesis by the first-order predicate logical expression by applying it to the knowledge data represented by. When,
Ru is the execution, program.

(Appendix 10)
The program described in Appendix 9
(C) When an inquiry by a first-order predicate logic expression including a term having no value is received, an observation corresponding to the inquiry is output according to the rule used by the observation model or a rule different from the rule. When,
(D) When the observation acquisition plan is input in the step (c), the observations output by the observation acquisition plan are acquired, and the acquired observations and the sample observations are used for hypothesis inference. Steps and steps to generate results,
(E) A step of updating the observation acquisition plan using the cost included in the hypothesis inference result generated in the step (d).
A program characterized by further executing.

(Appendix 11)
The program described in Appendix 10
Step (d) is executed until the cost of constructing the generated hypothetical inference result is less than or equal to a preset threshold.
A program characterized by that.

(Appendix 12)
The program described in any of the appendices 9 to 11 and
In the step (a), the observation with the cost is acquired in accordance with the rule stipulating that the observation with the lowest cost in the observation acquisition plan is applied to the observation model.
A program characterized by that.

Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made within the scope of the present invention in terms of the structure and details of the present invention.

As described above, according to the present invention, it is possible to suppress the redundancy of the acquisition of observations necessary for the verification of the hypothesis inference result. The present invention is useful in various fields in which hypothetical reasoning is utilized.

10 Inference knowledge storage unit 20 Observation acquisition plan storage unit 30 Observation acquisition unit 40 Hypothesis generation unit 50 Observation model 60 Observation output unit 70 Simulation execution unit 80 Observation acquisition plan update unit 100 Hypothesis generation device 110 Computer 111 CPU
112 Main memory 113 Storage device 114 Input interface 115 Display controller 116 Data reader / writer 117 Communication interface 118 Input device 119 Display device 120 Recording medium 121 Bus

Claims (6)

When a query by a first-order predicate logical expression including a term without a value is accepted, the observation model that outputs the observation corresponding to the query according to the set rule,
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
An observation acquisition unit that acquires the observation to which the cost is given, and
Knowledge of accepting and accepting observations that are first-order predicate logical expressions in which all terms have values, and the observations acquired by the observation acquisition unit, expressed by first-order predicate logical expressions including logical inclusion. A hypothesis generation unit that generates hypothetical inference results, which is composed of a first-order predicate logical expression that is applied to data and leads to a possible outcome, and a cost that indicates the evaluation of the hypothesis by the first-order predicate logical expression.
Is equipped with
A hypothesis generator characterized by this. The hypothesis generator according to claim 1.
When an inquiry by a first-order predicate logical expression including a term having no value is received, an observation output unit that outputs an observation corresponding to the inquiry according to the rule used by the observation model or a rule different from the rule. ,
The observation acquisition unit is made to input the observation acquisition plan to the observation output unit and the observations output by the input are acquired, and the hypothesis generation unit is made to acquire the observations and samples. A simulation execution unit that gives observations and generates hypothesis inference results,
An observation acquisition plan update unit that updates the observation acquisition plan using the cost included in the hypothesis inference result generated by the simulation execution unit.
A hypothesis generator characterized by further comprising. The hypothesis generator according to claim 2.
The simulation execution unit causes the observation acquisition unit to acquire observations and causes the hypothesis generation unit to generate hypothesis inference results until the cost of constructing the generated hypothesis inference result becomes equal to or less than a preset threshold value.
A hypothesis generator characterized by this. The hypothesis generator according to any one of claims 1 to 3.
The observation acquisition unit acquires the observation with the cost in accordance with the rule stipulating that the observation with the lowest cost in the observation acquisition plan is applied to the observation model.
A hypothesis generator characterized by this. (A) When an inquiry by a first-order predicate logical expression including a term having no value is accepted, an observation model corresponding to the inquiry is output according to the set rule.
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
To obtain the observation with the cost, and
(B) Accepting an observation that is a first-order predicate logical expression in which all terms have values, and the observation obtained in the step (a) above, the first-order predicate logical expression including logical inclusion. A step of generating a hypothetical inference result composed of a first-order predicate logical expression leading to a possible consequence and a cost indicating evaluation of a hypothesis by the first-order predicate logical expression by applying it to the knowledge data represented by. When,
Have,
A hypothesis generation method characterized by this. On the computer
(A) When an inquiry by a first-order predicate logical expression including a term having no value is accepted, an observation model corresponding to the inquiry is output according to the set rule.
Enter the observation acquisition plan, which is expressed by the first-order predicate logic formula and registers the observation to be queried and the cost to be obtained in the future.
To obtain the observation with the cost, and
(B) Accepting an observation that is a first-order predicate logical expression in which all terms have values, and the observation obtained in the step (a) above, the first-order predicate logical expression including logical inclusion. A step of generating a hypothetical inference result composed of a first-order predicate logical expression leading to a possible consequence and a cost indicating evaluation of a hypothesis by the first-order predicate logical expression by applying it to the knowledge data represented by. When,
Ru is the execution, program.

Images